Arguably, the role of a postgraduate supervisor is aligned to support the student to think critically and be able to write academically, but not necessarily as a “proofreader” for checking spelling, grammar, or syntax inconsistencies. This process can be realised as an AI-based tool providing near–real-time automated feedback to students, whilst reducing academic workload pressures. This paper proposes a formal (referring to rules) based proofreading process for information extraction, recognition, and correction of meaningful text chunks, images or figures, and tables (referred to as document tokens) from research proposals. The ultimate objective is custom-fit automated generation of feedback across multiple sections of submitted research proposals for a large number of authors. To this effect, an AI-driven tool was developed to support the proofreading of research proposals by delivering timely, automated feedback at scale, named RX Autoproofreader. The tool’s test on eighty (80) publicly available research documents analysed for constituent parts recognition and proofreading in an algorithm programmed in a c# Visual Studio environment, which ran in 182 min, produced an average accuracy of 68%. During the test, the tool extracted and proofread a total of 14,231 document pages into meaningful items of title, author, supervisor, research date, aim, objectives, figures, tables, paragraphs, lists, headings, and page numbers. The recognition and extraction of unstructured texts into meaningful items is a document understanding task necessary for proofreading and feedback generation. It aims to complement existing spell and grammar tools for proofreading academic manuscripts.
Ade-Ibijola et al. (Mon,) studied this question.